NEWS

Art; distinguishing original from forgery

18 October 2022 Project in the spotlights

The average thrift store, attic and flea market is full of them; replicas of famous paintings. From sunflowers by Vincent van Gogh, to the milkmaid by Johannes Vermeer. The art recognition tool developed within the Art Recognition project is capable of evaluating the authenticity of such masterpieces. The system, based on AI technologies, analyzes their photographic reproduction. Professor Eric Postma of Tilburg University has been engaged in this development for more than 20 years.

The tool - developed as part of the EuroStars project in cooperation with the Swiss company ArtRecognition - is fast, reliable and unhampered by human preferences. Moreover, it can be used for many works of art simultaneously. Offering this program within the art world contributes to the integrity and transparency of the art market. 

Recognition of data 
When asked about how exactly the system works, Postma is happy to explain using paintings by Van Gogh. "If you have a lot of examples of Van Gogh paintings and also numerous forgeries, you can train a computer using 'machine learning.' The system takes in all the examples and learns to recognize the subtle visual differences between the real work and the copied variants." 

When you use the tool, a small "recognition cube" passes over the scanned-in painting, he continues. "That cube studies every piece of the artwork, so to speak. The computer extracts visual information from this and translates it into numbers. Those numbers the algorithm uses to determine whether it's a real Van Gogh or not." 

Back in time 
The idea for using artificial intelligence to analyze works of art actually originated among art. During a neo-impressionism exhibition around the year 2000, with many works by Vincent van Gogh to be exact. Postma was still working at Maastricht University at the time, as a professor of Artificial Intelligence. The professor had always been interested in perception and pattern recognition. "Already in my youth I was concerned with how it is possible that you perceive and see things. I wanted to know how that works in people and whether you can mimic or understand that process with a computer." 

From then on, Postma focused on developing visual recognition and classification techniques for cultural heritage. Not much later, he started developing digital analysis methods for paintings, the beginning of the art analysis project. Just after the grant application for the NWO (Netherlands Organization for Scientific Research) was out the door, his research group already made the New York Times in 2004. "Assessing paintings by computers instead of experts turned out to be a 'hot topic,'" he tells us now.

The future of artificial intelligence 
The question of whether AI will eventually take over human work is one that Postma gets regularly. He emphasizes the interaction between computers and humans: "Machines are better at recognizing patterns than humans. Moreover, they don't get tired or distracted. But humans can outline context; something systems cannot yet do ( by a long shot)." As an example, the professor cites a painter's time image. Something a computer doesn't know, but the art expert does. "Meticulous tasks, such as studying brush strokes or - when we look outside the art world - an X-ray, AI technology can take over just fine. But this remains hand-in-hand with the human brain being able to put the findings into context." 

The X-ray, by the way, is a good example of how to utilize Postma's technology as well. Or think of an industrial site that you want to monitor with a security camera. Here, too, you can train the system on safe and unsafe situations. The condition is that you have enough data at your disposal to train the AI algorithms. 

Swiss organization takes over tool 
"Through MindLabs, we get in touch with companies and institutes that provide data for our recognition tool," Postma explains the cooperation with the ecosystem. In addition, Postma started a collaboration with a Swiss entrepreneur in 2020. "I received almost weekly requests from individuals if I could analyze a painting. And whether I didn't want to translate my research to a company offering that analysis. But the commercial slant you need for that is not in my genes. It doesn't interest me that much." Postma is now attached to the Swiss company ArtRecognition as a scientific advisor. Students at Tilburg University get the opportunity to do research within that organization. "A win-win situation." 

Since working with the Swiss company, Postma has more time to focus his attention on other things. And after 20 years in art recognition, he likes that he can re-enter the AI research field with a broader perspective. But Postma's focus remains with his main interest: "Pattern recognition is and will remain the thing that fascinates me the most."